1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26
27 #include <sys/dtrace.h>
28 #include <sys/systrace.h>
29 #include <sys/stat.h>
30 #include <sys/systm.h>
31 #include <sys/conf.h>
32 #include <sys/ddi.h>
33 #include <sys/sunddi.h>
34 #include <sys/atomic.h>
35
36 #define SYSTRACE_ARTIFICIAL_FRAMES 1
37
38 #define SYSTRACE_SHIFT 16
39 #define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT)
40 #define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1))
41 #define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id))
42 #define SYSTRACE_RETURN(id) (id)
43
44 #if ((1 << SYSTRACE_SHIFT) <= NSYSCALL)
45 #error 1 << SYSTRACE_SHIFT must exceed number of system calls
46 #endif
47
48 static dev_info_t *systrace_devi;
49 static dtrace_provider_id_t systrace_id;
50
51 static void
52 systrace_init(struct sysent *actual, systrace_sysent_t **interposed)
53 {
54 systrace_sysent_t *sysent = *interposed;
55 int i;
56
57 if (sysent == NULL) {
58 *interposed = sysent = kmem_zalloc(sizeof (systrace_sysent_t) *
59 NSYSCALL, KM_SLEEP);
60 }
61
62 for (i = 0; i < NSYSCALL; i++) {
63 struct sysent *a = &actual[i];
64 systrace_sysent_t *s = &sysent[i];
65
66 if (LOADABLE_SYSCALL(a) && !LOADED_SYSCALL(a))
67 continue;
68
69 if (a->sy_callc == dtrace_systrace_syscall)
70 continue;
71
72 #ifdef _SYSCALL32_IMPL
73 if (a->sy_callc == dtrace_systrace_syscall32)
74 continue;
75 #endif
76
77 s->stsy_underlying = a->sy_callc;
78 }
79 }
80
81 /*ARGSUSED*/
82 static void
83 systrace_provide(void *arg, const dtrace_probedesc_t *desc)
84 {
85 int i;
86
87 if (desc != NULL)
88 return;
89
90 systrace_init(sysent, &systrace_sysent);
91 #ifdef _SYSCALL32_IMPL
92 systrace_init(sysent32, &systrace_sysent32);
93 #endif
94
95 for (i = 0; i < NSYSCALL; i++) {
96 if (systrace_sysent[i].stsy_underlying == NULL)
97 continue;
98
99 if (dtrace_probe_lookup(systrace_id, NULL,
100 syscallnames[i], "entry") != 0)
101 continue;
102
103 (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i],
104 "entry", SYSTRACE_ARTIFICIAL_FRAMES,
105 (void *)((uintptr_t)SYSTRACE_ENTRY(i)));
106 (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i],
107 "return", SYSTRACE_ARTIFICIAL_FRAMES,
108 (void *)((uintptr_t)SYSTRACE_RETURN(i)));
109
110 systrace_sysent[i].stsy_entry = DTRACE_IDNONE;
111 systrace_sysent[i].stsy_return = DTRACE_IDNONE;
112 #ifdef _SYSCALL32_IMPL
113 systrace_sysent32[i].stsy_entry = DTRACE_IDNONE;
114 systrace_sysent32[i].stsy_return = DTRACE_IDNONE;
115 #endif
116 }
117 }
118
119 /*ARGSUSED*/
120 static void
121 systrace_destroy(void *arg, dtrace_id_t id, void *parg)
122 {
123 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
124
125 /*
126 * There's nothing to do here but assert that we have actually been
127 * disabled.
128 */
129 if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
130 ASSERT(systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE);
131 #ifdef _SYSCALL32_IMPL
132 ASSERT(systrace_sysent32[sysnum].stsy_entry == DTRACE_IDNONE);
133 #endif
134 } else {
135 ASSERT(systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE);
136 #ifdef _SYSCALL32_IMPL
137 ASSERT(systrace_sysent32[sysnum].stsy_return == DTRACE_IDNONE);
138 #endif
139 }
140 }
141
142 /*ARGSUSED*/
143 static int
144 systrace_enable(void *arg, dtrace_id_t id, void *parg)
145 {
146 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
147 int enabled = (systrace_sysent[sysnum].stsy_entry != DTRACE_IDNONE ||
148 systrace_sysent[sysnum].stsy_return != DTRACE_IDNONE);
149
150 if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
151 systrace_sysent[sysnum].stsy_entry = id;
152 #ifdef _SYSCALL32_IMPL
153 systrace_sysent32[sysnum].stsy_entry = id;
154 #endif
155 } else {
156 systrace_sysent[sysnum].stsy_return = id;
157 #ifdef _SYSCALL32_IMPL
158 systrace_sysent32[sysnum].stsy_return = id;
159 #endif
160 }
161
162 if (enabled) {
163 ASSERT(sysent[sysnum].sy_callc == dtrace_systrace_syscall);
164 return (0);
165 }
166
167 (void) atomic_cas_ptr(&sysent[sysnum].sy_callc,
168 (void *)systrace_sysent[sysnum].stsy_underlying,
169 (void *)dtrace_systrace_syscall);
170 #ifdef _SYSCALL32_IMPL
171 (void) atomic_cas_ptr(&sysent32[sysnum].sy_callc,
172 (void *)systrace_sysent32[sysnum].stsy_underlying,
173 (void *)dtrace_systrace_syscall32);
174 #endif
175 return (0);
176 }
177
178 /*ARGSUSED*/
179 static void
180 systrace_disable(void *arg, dtrace_id_t id, void *parg)
181 {
182 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
183 int disable = (systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE ||
184 systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE);
185
186 if (disable) {
187 (void) atomic_cas_ptr(&sysent[sysnum].sy_callc,
188 (void *)dtrace_systrace_syscall,
189 (void *)systrace_sysent[sysnum].stsy_underlying);
190
191 #ifdef _SYSCALL32_IMPL
192 (void) atomic_cas_ptr(&sysent32[sysnum].sy_callc,
193 (void *)dtrace_systrace_syscall32,
194 (void *)systrace_sysent32[sysnum].stsy_underlying);
195 #endif
196 }
197
198 if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
199 systrace_sysent[sysnum].stsy_entry = DTRACE_IDNONE;
200 #ifdef _SYSCALL32_IMPL
201 systrace_sysent32[sysnum].stsy_entry = DTRACE_IDNONE;
202 #endif
203 } else {
204 systrace_sysent[sysnum].stsy_return = DTRACE_IDNONE;
205 #ifdef _SYSCALL32_IMPL
206 systrace_sysent32[sysnum].stsy_return = DTRACE_IDNONE;
207 #endif
208 }
209 }
210
211 static dtrace_pattr_t systrace_attr = {
212 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
213 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
214 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
215 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
216 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
217 };
218
219 static dtrace_pops_t systrace_pops = {
220 systrace_provide,
221 NULL,
222 systrace_enable,
223 systrace_disable,
224 NULL,
225 NULL,
226 NULL,
227 NULL,
228 NULL,
229 systrace_destroy
230 };
231
232 static int
233 systrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
234 {
235 switch (cmd) {
236 case DDI_ATTACH:
237 break;
238 case DDI_RESUME:
239 return (DDI_SUCCESS);
240 default:
241 return (DDI_FAILURE);
242 }
243
244 systrace_probe = (void (*)())dtrace_probe;
245 membar_enter();
246
247 if (ddi_create_minor_node(devi, "systrace", S_IFCHR, 0,
248 DDI_PSEUDO, NULL) == DDI_FAILURE ||
249 dtrace_register("syscall", &systrace_attr, DTRACE_PRIV_USER, NULL,
250 &systrace_pops, NULL, &systrace_id) != 0) {
251 systrace_probe = systrace_stub;
252 ddi_remove_minor_node(devi, NULL);
253 return (DDI_FAILURE);
254 }
255
256 ddi_report_dev(devi);
257 systrace_devi = devi;
258
259 return (DDI_SUCCESS);
260 }
261
262 static int
263 systrace_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
264 {
265 switch (cmd) {
266 case DDI_DETACH:
267 break;
268 case DDI_SUSPEND:
269 return (DDI_SUCCESS);
270 default:
271 return (DDI_FAILURE);
272 }
273
274 if (dtrace_unregister(systrace_id) != 0)
275 return (DDI_FAILURE);
276
277 ddi_remove_minor_node(devi, NULL);
278 systrace_probe = systrace_stub;
279 return (DDI_SUCCESS);
280 }
281
282 /*ARGSUSED*/
283 static int
284 systrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
285 {
286 int error;
287
288 switch (infocmd) {
289 case DDI_INFO_DEVT2DEVINFO:
290 *result = (void *)systrace_devi;
291 error = DDI_SUCCESS;
292 break;
293 case DDI_INFO_DEVT2INSTANCE:
294 *result = (void *)0;
295 error = DDI_SUCCESS;
296 break;
297 default:
298 error = DDI_FAILURE;
299 }
300 return (error);
301 }
302
303 /*ARGSUSED*/
304 static int
305 systrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
306 {
307 return (0);
308 }
309
310 static struct cb_ops systrace_cb_ops = {
311 systrace_open, /* open */
312 nodev, /* close */
313 nulldev, /* strategy */
314 nulldev, /* print */
315 nodev, /* dump */
316 nodev, /* read */
317 nodev, /* write */
318 nodev, /* ioctl */
319 nodev, /* devmap */
320 nodev, /* mmap */
321 nodev, /* segmap */
322 nochpoll, /* poll */
323 ddi_prop_op, /* cb_prop_op */
324 0, /* streamtab */
325 D_NEW | D_MP /* Driver compatibility flag */
326 };
327
328 static struct dev_ops systrace_ops = {
329 DEVO_REV, /* devo_rev, */
330 0, /* refcnt */
331 systrace_info, /* get_dev_info */
332 nulldev, /* identify */
333 nulldev, /* probe */
334 systrace_attach, /* attach */
335 systrace_detach, /* detach */
336 nodev, /* reset */
337 &systrace_cb_ops, /* driver operations */
338 NULL, /* bus operations */
339 nodev, /* dev power */
340 ddi_quiesce_not_needed, /* quiesce */
341 };
342
343 /*
344 * Module linkage information for the kernel.
345 */
346 static struct modldrv modldrv = {
347 &mod_driverops, /* module type (this is a pseudo driver) */
348 "System Call Tracing", /* name of module */
349 &systrace_ops, /* driver ops */
350 };
351
352 static struct modlinkage modlinkage = {
353 MODREV_1,
354 { (void *)&modldrv, NULL }
355 };
356
357 int
358 _init(void)
359 {
360 return (mod_install(&modlinkage));
361 }
362
363 int
364 _info(struct modinfo *modinfop)
365 {
366 return (mod_info(&modlinkage, modinfop));
367 }
368
369 int
370 _fini(void)
371 {
372 return (mod_remove(&modlinkage));
373 }